Detection of document modification based on deep neural networks

In this paper, we focus on the detection of the semantic and structural modifications in documents. We define the following six inter-document relations that we use to represent document modification: Eliminate, Extend, Merge, Split, Rewrite, and Reorder. We also develop a detection model based on a deep neural network to identify the relations between two given documents. We assumed that several modifications can be applied to a document; in this situation, the modifications can overlap each other, so it can be very difficult to detect the applied modifications. We represent a document pair by using a sentence-based similarity matrix, and the inter-document relations are then detected by applying the deep neural network to the similarity matrix. The experiments show that our model performed impressively in the detection of document modifications.